Spaces:
Running
Running
Update
Browse files
app.py
CHANGED
@@ -12,8 +12,6 @@ from transformers import AutoProcessor, Blip2ForConditionalGeneration
|
|
12 |
|
13 |
DESCRIPTION = "# [BLIP-2](https://github.com/salesforce/LAVIS/tree/main/projects/blip2)"
|
14 |
|
15 |
-
if (SPACE_ID := os.getenv("SPACE_ID")) is not None:
|
16 |
-
DESCRIPTION += f'\n<p>For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings. <a href="https://huggingface.co/spaces/{SPACE_ID}?duplicate=true"><img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space" /></a></p>'
|
17 |
if not torch.cuda.is_available():
|
18 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
19 |
|
@@ -21,40 +19,23 @@ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
|
21 |
|
22 |
MODEL_ID_OPT_6_7B = "Salesforce/blip2-opt-6.7b"
|
23 |
MODEL_ID_FLAN_T5_XXL = "Salesforce/blip2-flan-t5-xxl"
|
|
|
24 |
|
25 |
if torch.cuda.is_available():
|
26 |
-
|
27 |
-
|
28 |
-
# 'processor':
|
29 |
-
# AutoProcessor.from_pretrained(MODEL_ID_OPT_6_7B),
|
30 |
-
# 'model':
|
31 |
-
# Blip2ForConditionalGeneration.from_pretrained(MODEL_ID_OPT_6_7B,
|
32 |
-
# device_map='auto',
|
33 |
-
# load_in_8bit=True),
|
34 |
-
# },
|
35 |
-
MODEL_ID_FLAN_T5_XXL: {
|
36 |
-
"processor": AutoProcessor.from_pretrained(MODEL_ID_FLAN_T5_XXL),
|
37 |
-
"model": Blip2ForConditionalGeneration.from_pretrained(
|
38 |
-
MODEL_ID_FLAN_T5_XXL, device_map="auto", load_in_8bit=True
|
39 |
-
),
|
40 |
-
}
|
41 |
-
}
|
42 |
else:
|
43 |
-
|
|
|
44 |
|
45 |
|
46 |
def generate_caption(
|
47 |
-
model_id: str,
|
48 |
image: PIL.Image.Image,
|
49 |
decoding_method: str,
|
50 |
temperature: float,
|
51 |
length_penalty: float,
|
52 |
repetition_penalty: float,
|
53 |
) -> str:
|
54 |
-
model_info = model_dict[model_id]
|
55 |
-
processor = model_info["processor"]
|
56 |
-
model = model_info["model"]
|
57 |
-
|
58 |
inputs = processor(images=image, return_tensors="pt").to(device, torch.float16)
|
59 |
generated_ids = model.generate(
|
60 |
pixel_values=inputs.pixel_values,
|
@@ -72,7 +53,6 @@ def generate_caption(
|
|
72 |
|
73 |
|
74 |
def answer_question(
|
75 |
-
model_id: str,
|
76 |
image: PIL.Image.Image,
|
77 |
text: str,
|
78 |
decoding_method: str,
|
@@ -80,10 +60,6 @@ def answer_question(
|
|
80 |
length_penalty: float,
|
81 |
repetition_penalty: float,
|
82 |
) -> str:
|
83 |
-
model_info = model_dict[model_id]
|
84 |
-
processor = model_info["processor"]
|
85 |
-
model = model_info["model"]
|
86 |
-
|
87 |
inputs = processor(images=image, text=text, return_tensors="pt").to(device, torch.float16)
|
88 |
generated_ids = model.generate(
|
89 |
**inputs,
|
@@ -107,7 +83,6 @@ def postprocess_output(output: str) -> str:
|
|
107 |
|
108 |
|
109 |
def chat(
|
110 |
-
model_id: str,
|
111 |
image: PIL.Image.Image,
|
112 |
text: str,
|
113 |
decoding_method: str,
|
@@ -123,7 +98,6 @@ def chat(
|
|
123 |
prompt = " ".join(history_qa)
|
124 |
|
125 |
output = answer_question(
|
126 |
-
model_id,
|
127 |
image,
|
128 |
prompt,
|
129 |
decoding_method,
|
@@ -164,24 +138,14 @@ examples = [
|
|
164 |
|
165 |
with gr.Blocks(css="style.css") as demo:
|
166 |
gr.Markdown(DESCRIPTION)
|
|
|
|
|
|
|
|
|
|
|
167 |
|
168 |
image = gr.Image(type="pil")
|
169 |
with gr.Accordion(label="Advanced settings", open=False):
|
170 |
-
with gr.Row():
|
171 |
-
model_id_caption = gr.Dropdown(
|
172 |
-
label="Model ID for image captioning",
|
173 |
-
choices=[MODEL_ID_OPT_6_7B, MODEL_ID_FLAN_T5_XXL],
|
174 |
-
value=MODEL_ID_FLAN_T5_XXL,
|
175 |
-
interactive=False,
|
176 |
-
visible=False,
|
177 |
-
)
|
178 |
-
model_id_chat = gr.Dropdown(
|
179 |
-
label="Model ID for VQA",
|
180 |
-
choices=[MODEL_ID_OPT_6_7B, MODEL_ID_FLAN_T5_XXL],
|
181 |
-
value=MODEL_ID_FLAN_T5_XXL,
|
182 |
-
interactive=False,
|
183 |
-
visible=False,
|
184 |
-
)
|
185 |
sampling_method = gr.Radio(
|
186 |
label="Text Decoding Method",
|
187 |
choices=["Beam search", "Nucleus sampling"],
|
@@ -225,16 +189,12 @@ with gr.Blocks(css="style.css") as demo:
|
|
225 |
|
226 |
gr.Examples(
|
227 |
examples=examples,
|
228 |
-
inputs=[
|
229 |
-
image,
|
230 |
-
vqa_input,
|
231 |
-
],
|
232 |
)
|
233 |
|
234 |
caption_button.click(
|
235 |
fn=generate_caption,
|
236 |
inputs=[
|
237 |
-
model_id_caption,
|
238 |
image,
|
239 |
sampling_method,
|
240 |
temperature,
|
@@ -246,7 +206,6 @@ with gr.Blocks(css="style.css") as demo:
|
|
246 |
)
|
247 |
|
248 |
chat_inputs = [
|
249 |
-
model_id_chat,
|
250 |
image,
|
251 |
vqa_input,
|
252 |
sampling_method,
|
@@ -296,4 +255,5 @@ with gr.Blocks(css="style.css") as demo:
|
|
296 |
queue=False,
|
297 |
)
|
298 |
|
299 |
-
|
|
|
|
12 |
|
13 |
DESCRIPTION = "# [BLIP-2](https://github.com/salesforce/LAVIS/tree/main/projects/blip2)"
|
14 |
|
|
|
|
|
15 |
if not torch.cuda.is_available():
|
16 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
17 |
|
|
|
19 |
|
20 |
MODEL_ID_OPT_6_7B = "Salesforce/blip2-opt-6.7b"
|
21 |
MODEL_ID_FLAN_T5_XXL = "Salesforce/blip2-flan-t5-xxl"
|
22 |
+
MODEL_ID = os.getenv("MODEL_ID", MODEL_ID_FLAN_T5_XXL)
|
23 |
|
24 |
if torch.cuda.is_available():
|
25 |
+
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
26 |
+
model = Blip2ForConditionalGeneration.from_pretrained(MODEL_ID, device_map="auto", load_in_8bit=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
else:
|
28 |
+
processor = None
|
29 |
+
model = None
|
30 |
|
31 |
|
32 |
def generate_caption(
|
|
|
33 |
image: PIL.Image.Image,
|
34 |
decoding_method: str,
|
35 |
temperature: float,
|
36 |
length_penalty: float,
|
37 |
repetition_penalty: float,
|
38 |
) -> str:
|
|
|
|
|
|
|
|
|
39 |
inputs = processor(images=image, return_tensors="pt").to(device, torch.float16)
|
40 |
generated_ids = model.generate(
|
41 |
pixel_values=inputs.pixel_values,
|
|
|
53 |
|
54 |
|
55 |
def answer_question(
|
|
|
56 |
image: PIL.Image.Image,
|
57 |
text: str,
|
58 |
decoding_method: str,
|
|
|
60 |
length_penalty: float,
|
61 |
repetition_penalty: float,
|
62 |
) -> str:
|
|
|
|
|
|
|
|
|
63 |
inputs = processor(images=image, text=text, return_tensors="pt").to(device, torch.float16)
|
64 |
generated_ids = model.generate(
|
65 |
**inputs,
|
|
|
83 |
|
84 |
|
85 |
def chat(
|
|
|
86 |
image: PIL.Image.Image,
|
87 |
text: str,
|
88 |
decoding_method: str,
|
|
|
98 |
prompt = " ".join(history_qa)
|
99 |
|
100 |
output = answer_question(
|
|
|
101 |
image,
|
102 |
prompt,
|
103 |
decoding_method,
|
|
|
138 |
|
139 |
with gr.Blocks(css="style.css") as demo:
|
140 |
gr.Markdown(DESCRIPTION)
|
141 |
+
gr.DuplicateButton(
|
142 |
+
value="Duplicate Space for private use",
|
143 |
+
elem_id="duplicate-button",
|
144 |
+
visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
|
145 |
+
)
|
146 |
|
147 |
image = gr.Image(type="pil")
|
148 |
with gr.Accordion(label="Advanced settings", open=False):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
149 |
sampling_method = gr.Radio(
|
150 |
label="Text Decoding Method",
|
151 |
choices=["Beam search", "Nucleus sampling"],
|
|
|
189 |
|
190 |
gr.Examples(
|
191 |
examples=examples,
|
192 |
+
inputs=[image, vqa_input],
|
|
|
|
|
|
|
193 |
)
|
194 |
|
195 |
caption_button.click(
|
196 |
fn=generate_caption,
|
197 |
inputs=[
|
|
|
198 |
image,
|
199 |
sampling_method,
|
200 |
temperature,
|
|
|
206 |
)
|
207 |
|
208 |
chat_inputs = [
|
|
|
209 |
image,
|
210 |
vqa_input,
|
211 |
sampling_method,
|
|
|
255 |
queue=False,
|
256 |
)
|
257 |
|
258 |
+
if __name__ == "__main__":
|
259 |
+
demo.queue(max_size=10).launch()
|
style.css
CHANGED
@@ -1,3 +1,10 @@
|
|
1 |
h1 {
|
2 |
text-align: center;
|
3 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
h1 {
|
2 |
text-align: center;
|
3 |
}
|
4 |
+
|
5 |
+
#duplicate-button {
|
6 |
+
margin: auto;
|
7 |
+
color: #fff;
|
8 |
+
background: #1565c0;
|
9 |
+
border-radius: 100vh;
|
10 |
+
}
|